Support ipv6 in http client #310
Conversation
HanLee13
commented
Dec 20, 2023
HanLee13
commented
- Support ipv6 in http client.
- refactor KernelSocketClient by removing socket_family in KernelSocketClient.
- apply dns load balance in DefaultResolver.
- fix bug in DefaultResolver when timeout happen.
- Add big writes/uid/gid/du support to FUSE
- In Fuse 2.x, without big writes, write io size is limited to 4k. Turing it on can enlarge write io size to 128K at maximum.
- Support du by adding st_blocks to xmp_getattr result.
- Add default uid/gid to xmp_getattr to avoid permission problems.
- refactor KernelSocketClient. - apply dns load balance in DefaultResolver. - fix bug in DefaultResolver when timeout happen.
- In Fuse 2.x, without big writes, write io size is limited to 4k. Turing it on can enlarge write io size to 128K at maximum. - Support du by adding st_blocks to xmp_getattr result. - Add default uid/gid to xmp_getattr to avoid permission problems.
|
|
net/utils.cpp
Outdated
| ipaddr.recycle(true); | ||
| if (ip.undefined() || ipaddr->empty()) ipaddr.recycle(true); | ||
| else { | ||
| for (auto it = ipaddr->begin(); it != ipaddr->end(); ++it) { |
There was a problem hiding this comment.
try std::list::remove(const T& value)
| return new IPAddr(ip); | ||
| } | ||
| return new IPAddr; // undefined addr | ||
| sem.wait(1, resolve_timeout_); |
There was a problem hiding this comment.
if time out, sem will be destructed upon return, but it may be referenced afterwards in the resolving thread.
There was a problem hiding this comment.
addrs may also be operated (erased or deleted) when the resolving thread is still running.
net/utils.cpp
Outdated
| gethostbyname(host, *addrs); | ||
| auto time_elapsed = std::chrono::duration_cast<std::chrono::microseconds>( | ||
| std::chrono::steady_clock::now() - now).count(); | ||
| if ((uint64_t)time_elapsed <= resolve_timeout_) sem.signal(1); |
There was a problem hiding this comment.
timing control may not be precise enough?
There was a problem hiding this comment.
Cause the photon thread should sleep for more than resolve_timeout_, this is enought to run correctly, right?
net/utils.cpp
Outdated
| auto ctr = [&]() -> IPAddr * { | ||
| std::vector<IPAddr> addrs; | ||
| auto ctr = [&]() -> std::list<IPAddr>* { | ||
| auto addrs = new std::list<IPAddr>(); |
There was a problem hiding this comment.
If using unordered_set, get element in a round-robin order would be O(N).
net/utils.cpp
Outdated
| if (ips->empty()) LOG_ERRNO_RETURN(0, IPAddr(), "Domain resolution for ` failed", host); | ||
| auto ret = ips->front(); | ||
| if (ips->size() > 1) { // access in round robin order | ||
| ips->pop_front(); |
There was a problem hiding this comment.
It's costly to pop and push every time. It's better to use intrusive_list.
net/utils.cpp
Outdated
| auto ctr = [&]() -> IPAddr * { | ||
| std::vector<IPAddr> addrs; | ||
| auto ctr = [&]() -> std::list<IPAddr>* { | ||
| auto addrs = new std::list<IPAddr>(); |
HanLee13
force-pushed
the
release/0.7
branch
2 times, most recently
from
789514d
to
450dd57
Compare
| auto ctr = [&]() -> IPAddr * { | ||
| std::vector<IPAddr> addrs; | ||
| auto ctr = [&]() -> IPAddrList* { | ||
| auto addrs = new IPAddrList(); |
There was a problem hiding this comment.
The list doesn't need to be new-ed, as it's only a pointer to the node, initially nullptr. And nodes are directly pushed-back into the list.
There was a problem hiding this comment.
ObjectCache need its value be a pointer, otherwise using ObjectCache::Borrow would be a problem.
net/utils.cpp
Outdated
| if (ips->empty()) LOG_ERRNO_RETURN(0, IPAddr(), "Domain resolution for ` failed", host); | ||
| auto ret = ips->front(); | ||
| if (!ret->single()) { // access in round robin order | ||
| auto front = ips->pop_front(); |
There was a problem hiding this comment.
Just update the pointer in the list object to point to the next node, without popping and pushing.
net/utils.cpp
Outdated
| ipaddr.recycle(true); | ||
| if (ip.undefined() || ipaddr->empty()) ipaddr.recycle(true); | ||
| else { | ||
| for (auto it : *ipaddr) { |
There was a problem hiding this comment.
It's probably better to iterate in reversed order, because the item to be discarded is likely at the end of the ring.
HanLee13
force-pushed
the
release/0.7
branch
from
450dd57
to
1c653cd
Compare
